Headless GPU Server 上启用Virgl实现虚机图形加速

2023-05-16 14:16| 来源: 网络整理| 查看: 265

1. 目的

验证在headless GPU Server上启用virgl加速的可行性

2. 环境

硬件: nvidia gtx1060 6G独显笔记本（附带intel集显）:

$ sudo lspci | grep -i vga 00:02.0 VGA compatible controller: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630] 01:00.0 VGA compatible controller: NVIDIA Corporation GP106M [GeForce GTX 1060 Mobile] (rev a1) $ sudo nvidia-smi Fri Dec 9 15:13:58 2022 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 525.60.11 Driver Version: 525.60.11 CUDA Version: 12.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 NVIDIA GeForce ... Off | 00000000:01:00.0 Off | N/A | | N/A 52C P8 7W / N/A | 203MiB / 6144MiB | 17% Default | | | | N/A | +-------------------------------+----------------------+----------------------+

系统及软件(arch下依赖qemu/qemu-ui-egl-headless):

$ uname -r 6.0.12-arch1-1 $ cat /etc/issue Arch Linux \r (\l) $ qemu-system-x86_64 --version QEMU emulator version 7.1.0 Copyright (c) 2003-2022 Fabrice Bellard and the QEMU Project developers $ sudo pacman -Q | grep -i qemu | grep egl qemu-ui-egl-headless 7.1.0-11

VM环境:

$ cat /etc/issue Ubuntu 20.04.5 LTS \n \l $ uname -r 5.15.0-56-generic 3. 启动VM

纯命令行下启动（无x11登陆）:

sudo qemu-system-x86_64 -name ubuntu2204 -drive file=`pwd`/ubuntudesktop2004_spice.qcow2,if=virtio -m 8192 -enable-kvm -cpu host -smp 8,sockets=1,cores=8,threads=1 -spice unix=on,addr=/tmp/spice.sock,disable-ticketing=on,plaintext-channel=default,seamless-migration=on,image-compression=off,jpeg-wan-compression=never,zlib-glz-wan-compression=never,streaming-video=off,playback-compression=off \ -netdev user,id=vnet,hostfwd=:127.0.0.1:2278-:22,hostfwd=tcp::5555-:5555,hostfwd=tcp::5556-:5556,hostfwd=tcp::5557-:5557,hostfwd=tcp::5800-:5800,hostfwd=tcp::3389-:3389,hostfwd=tcp::14000-:4000 -device virtio-net-pci,netdev=vnet \ -display egl-headless,rendernode=/dev/dri/renderD129 \ -device virtio-vga-gl,max_outputs=1 \ -device virtio-serial-pci \ -chardev spicevmc,id=charchannel1,name=vdagent \ -device virtserialport,chardev=charchannel1,id=channel1,name=com.redhat.spice.0

其中rendernode为nvidia gtx1060.

4. 验证

另起一终端，ssh到vm后验证:

$ export DISPLAY=:0 $ glxinfo | grep -i render direct rendering: Yes GLX_MESA_query_renderer, GLX_MESA_swap_control, GLX_NV_float_buffer, GLX_MESA_copy_sub_buffer, GLX_MESA_query_renderer, GLX_MESA_swap_control, Extended renderer info (GLX_MESA_query_renderer): Renderbuffer free memory - total: 21 MB, largest block: 21 MB Renderbuffer free aux. memory - total: 0 MB, largest block: 0 MB OpenGL renderer string: virgl GL_ARB_conditional_render_inverted, GL_ARB_conservative_depth, GL_NV_conditional_render, GL_NV_copy_image, GL_NV_depth_clamp, GL_ARB_compute_shader, GL_ARB_conditional_render_inverted, GL_NV_conditional_render, GL_NV_copy_image, GL_NV_depth_clamp, GL_EXT_render_snorm, GL_EXT_robustness, GL_EXT_sRGB_write_control, GL_NV_conditional_render, GL_NV_draw_buffers, GL_NV_fbo_color_attachments, GL_OES_element_index_uint, GL_OES_fbo_render_mipmap, $ ls /dev/dri/ by-path card0 renderD128 $ sudo dmesg | grep -i virgl [ 2.257506] [drm] features: +virgl +edid -resource_blob -host_visible $ glmark2 ======================================================= glmark2 2021.02 ======================================================= OpenGL Information GL_VENDOR: Mesa/X.org GL_RENDERER: virgl GL_VERSION: 3.1 Mesa 21.2.6 ======================================================= [build] use-vbo=false: FPS: 5102 FrameTime: 0.196 ms ......... [loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 6301 FrameTime: 0.159 ms [loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 5922 FrameTime: 0.169 ms [loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 5785 FrameTime: 0.173 ms ======================================================= glmark2 Score: 3996 =======================================================

最终跑分为接近4000分，由此可见headless server下virgl的2d/3d均可用, 且使用了nvidia卡的加速能力。

【本文地址】

公司简介

联系我们